Add decompose + orchestrate skills with operational improvements

New skills for container agent task orchestration:
- /decompose: break tasks into subtasks with dependency graph, write .agent-tasks.json
- /orchestrate: check task state, launch container agents in git worktrees, poll for completion

Updated with learnings from first real run (agent-runtimes M3, 9 tasks):
- Prompts must end with "run pytest and fix failures"
- State import conventions explicitly in prompts
- Warn about uncommitted files before decomposing (worktrees need committed content)
- Copy dependency outputs into downstream worktrees before launching

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

FUTURE.md

@@ -11,3 +11,9 @@
 - **Idea:** A `/new-milestone M9 "Velero backup"` skill that scaffolds the milestone
 - **Open questions:** How much should be templated vs. left to Claude?
 - **Depends on:** Stable milestone conventions across projects
+
+## F3: Task retry skill
+- **Problem:** When a container agent task fails (via `/orchestrate`), retrying requires manually editing `.agent-tasks.json` to reset status and optionally appending the error context to the prompt
+- **Idea:** A `/retry <task-id>` skill that reads the failed task's error/logs, appends them to the prompt as "Previous attempt failed with: ...", resets status to `pending`, and cleans up the old worktree/branch
+- **Open questions:** Should the retry prompt include the full error log or a summary? Should it increment a retry counter and give up after N attempts?
+- **Depends on:** `/decompose` and `/orchestrate` being validated in real use